Incorporating Intensity Features And Shape Features For Image Registration

Image registration is an important branch of computer vision and digital image processing.It has been widely used in medical, remote sensing, industrial analysis system, and many other fields. It is a crucial issue to perform registration with high precision, high efficiency and high robustness in the practical applications.To improve the accuracy and robustness two effective registration methods are proposed in the paper. One method is to combine image geometric features and intensity features to register images and the other method is to incorporate image intensity features and structural image representation for groupwise registration.1. Combining image geometric features and intensity features for image registration. First,features of the reference and floating images are extracted using scale invariant feature transform(SIFT). Second, SIFT features of the reference and floating images are matched and the affine transformation model based on the minimum mean square error(MMSE) method of matching feature points is modeled. Then, the rough registered image by transforming the floating image based on the affine transformation model is obtained. Finally, the individual entropy correlation coefficient(IECC) is used as the similarity measure to refine the rough registered image.Experiments show that the proposed method can effectively improve the accuracy and robustness of registration.2. Incorporating intensity and structural features for groupwise registration. First, the structural image representation of the original images is calculated. Second, the original images and the structural image representation are used as the input images, which are then transformed by affine transformation. The entropy is used as the similarity measure to optimize the best affinetransformation parameters to get the rough registration images. Finally, B-spline transformation is used to refine the rough registration images. Experiments show that the method can effectively improve the accuracy and robustness of registration.